MongoDB CDC Connector

The MongoDB CDC connector allows for reading snapshot data and incremental data from MongoDB. This document describes how to set up the MongoDB CDC connector to run SQL queries against MongoDB.

Dependencies

To set up the MongoDB CDC connector, the following table provides dependency information for both projects using a build automation tool (such as Maven or SBT) and SQL Client with SQL JAR bundles.

Maven dependency

<dependency>
    <groupId>org.apache.flink</groupId>
    <artifactId>flink-connector-mongodb-cdc</artifactId>
    <version>3.8-SNAPSHOT</version>
</dependency>

SQL Client JAR

Download link is available only for stable releases.

Download flink-sql-connector-mongodb-cdc and put it under <FLINK_HOME>/lib/.

Note: Refer to flink-sql-connector-mongodb-cdc, more released versions will be available in the Maven central warehouse.

Setup MongoDB

Availability

• MongoDB version

  MongoDB version >= 3.6
  We use change streams feature (new in version 3.6) to capture change data.

• Cluster Deployment

  replica sets or sharded clusters is required.

• Storage Engine

  WiredTiger storage engine is required.

• Replica set protocol version

  Replica set protocol version 1 (pv1) is required.
  Starting in version 4.0, MongoDB only supports pv1. pv1 is the default for all new replica sets created with MongoDB 3.2 or later.

• Privileges

  changeStream and read privileges are required by MongoDB Kafka Connector.

  You can use the following example for simple authorization.
  For more detailed authorization, please refer to MongoDB Database User Roles.

  use admin;
  db.createRole(
      {
          role: "flinkrole",
          privileges: [{
              // Grant privileges on all non-system collections in all databases
              resource: { db: "", collection: "" },
              actions: [
                  "splitVector",
                  "listDatabases",
                  "listCollections",
                  "collStats",
                  "find",
                  "changeStream" ]
          }],
          roles: [
              // Read config.collections and config.chunks
              // for sharded cluster snapshot splitting.
              { role: 'read', db: 'config' }
          ]
      }
  );
  
  db.createUser(
    {
        user: 'flinkuser',
        pwd: 'flinkpw',
        roles: [
           { role: 'flinkrole', db: 'admin' }
        ]
    }
  );

How to create a MongoDB CDC table

The MongoDB CDC table can be defined as following:

-- register a MongoDB table 'products' in Flink SQL
CREATE TABLE products (
  _id STRING, // must be declared
  name STRING,
  weight DECIMAL(10,3),
  tags ARRAY<STRING>, -- array
  price ROW<amount DECIMAL(10,2), currency STRING>, -- embedded document
  suppliers ARRAY<ROW<name STRING, address STRING>>, -- embedded documents
  PRIMARY KEY(_id) NOT ENFORCED
) WITH (
  'connector' = 'mongodb-cdc',
  'hosts' = 'localhost:27017,localhost:27018,localhost:27019',
  'username' = 'flinkuser',
  'password' = 'flinkpw',
  'database' = 'inventory',
  'collection' = 'products'
);

-- read snapshot and change events from products collection
SELECT * FROM products;

Note that

MongoDB's change event record doesn't have updated before message. So, we can only convert it to Flink's UPSERT changelog stream. An upsert stream requires a unique key, so we must declare _id as primary key. We can't declare other column as primary key, because delete operation does not contain the key and value besides _id and sharding key.

Connector Options

Option	Required	Default	Type	Description
connector	required	(none)	String	Specify what connector to use, here should be mongodb-cdc.
scheme	optional	mongodb	String	The protocol connected to MongoDB. eg. mongodb or mongodb+srv.
hosts	required	(none)	String	The comma-separated list of hostname and port pairs of the MongoDB servers. eg. localhost:27017,localhost:27018
username	optional	(none)	String	Name of the database user to be used when connecting to MongoDB. This is required only when MongoDB is configured to use authentication.
password	optional	(none)	String	Password to be used when connecting to MongoDB. This is required only when MongoDB is configured to use authentication.
database	optional	(none)	String	Name of the database to watch for changes. If not set then all databases will be captured. The database also supports regular expressions to monitor multiple databases matching the regular expression.
collection	optional	(none)	String	Name of the collection in the database to watch for changes. If not set then all collections will be captured. The collection also supports regular expressions to monitor multiple collections matching fully-qualified collection identifiers.
connection.options	optional	(none)	String	The ampersand-separated connection options of MongoDB. eg. replicaSet=test&connectTimeoutMS=300000
scan.startup.mode	optional	initial	String	Optional startup mode for MongoDB CDC consumer, valid enumerations are "initial", "latest-offset" and "timestamp". Please see Startup Reading Position section for more detailed information.
scan.startup.timestamp-millis	optional	(none)	Long	Timestamp in millis of the start point, only used for 'timestamp' startup mode.
initial.snapshotting.queue.size	optional	16000	Integer	The max size of the queue to use when copying data. Only available when scan.startup.mode is set to 'initial'. Note: The deprecated option name is copy.existing.queue.size. To be compatible with old versions of jobs, this parameter is still available, but it is recommended to upgrade to the new option name.
initial.snapshotting.max.threads	optional	Processors Count	Integer	The number of threads to use when performing the data copy. Only available when scan.startup.mode is set to 'initial'. Note: The deprecated option name is copy.existing.max.threads. To be compatible with old versions of jobs, this parameter is still available, but it is recommended to upgrade to the new option name.
initial.snapshotting.pipeline	optional	(none)	String	An array of JSON objects describing the pipeline operations to run when copying existing data. This can improve the use of indexes by the copying manager and make copying more efficient. eg. [{"$match": {"closed": "false"}}] ensures that only documents in which the closed field is set to false are copied. The initial.snapshotting.pipeline config is only available when scan.startup.mode is set to 'initial', and is only used in Debezium mode and cannot be used in incremental snapshot mode because the semantic is inconsistent. Note: The deprecated option name is copy.existing.pipeline. To be compatible with old versions of jobs, this parameter is still available, but it is recommended to upgrade to the new option name.
batch.size	optional	1024	Integer	The cursor batch size.
poll.max.batch.size	optional	1024	Integer	Maximum number of change stream documents to include in a single batch when polling for new data.
poll.await.time.ms	optional	1000	Integer	The amount of time to wait before checking for new results on the change stream.
heartbeat.interval.ms	optional	0	Integer	The length of time in milliseconds between sending heartbeat messages. Use 0 to disable.
scan.full-changelog	optional	false	Boolean	Whether try to generate full-mode changelog based on pre- and post-images in MongoDB. Refer to Full Changelog for more details. Supports MongoDB 6.0 and above only.
scan.incremental.snapshot.enabled	optional	false	Boolean	Whether enable incremental snapshot. The incremental snapshot feature only supports after MongoDB 4.0.
scan.incremental.snapshot.chunk.size.mb	optional	64	Integer	The chunk size mb of incremental snapshot.
scan.incremental.snapshot.chunk.samples	optional	20	Integer	The samples count per chunk when using sample partition strategy during incremental snapshot.
scan.incremental.close-idle-reader.enabled	optional	false	Boolean	Whether to close idle readers at the end of the snapshot phase. The flink version is required to be greater than or equal to 1.14 when 'execution.checkpointing.checkpoints-after-tasks-finish.enabled' is set to true. If the flink version is greater than or equal to 1.15, the default value of 'execution.checkpointing.checkpoints-after-tasks-finish.enabled' has been changed to true, so it does not need to be explicitly configured 'execution.checkpointing.checkpoints-after-tasks-finish.enabled' = 'true'
scan.incremental.snapshot.metadata.release.enabled	optional	false	Boolean	Whether to release the snapshot split metadata (assigned splits, finished offsets and table schemas) held by the source coordinator once the source has entered the stream phase, to reduce JobManager memory on jobs with a very large number of snapshot splits. Disabled by default. Incompatible with scan.newly-added-table.enabled: enabling both fails at startup, and a job that has released the metadata cannot later enable newly-added-table scanning. Release happens only after a successful checkpoint; if checkpointing is disabled or no checkpoint completes, the metadata is retained, so this option has no effect without checkpointing. A checkpoint or savepoint taken with this option enabled cannot be used to restore the job after downgrading to Flink CDC 3.6.0 or an earlier version.
scan.cursor.no-timeout	optional	true	Boolean	MongoDB server normally times out idle cursors after an inactivity period (10 minutes) to prevent excess memory use. Set this option to true to prevent that. Only available when parallelism snapshot is enabled.
scan.incremental.snapshot.unbounded-chunk-first.enabled	optional	true	Boolean	Whether to assign the unbounded chunks first during snapshot reading phase. This might help reduce the risk of the TaskManager experiencing an out-of-memory (OOM) error when taking a snapshot of the largest unbounded chunk.
scan.incremental.snapshot.backfill.skip	optional	false	Boolean	Whether to skip backfill in snapshot reading phase. If backfill is skipped, changes on captured tables during snapshot phase will be consumed later in change log reading phase instead of being merged into the snapshot. WARNING: Skipping backfill might lead to data inconsistency because some change log events happened within the snapshot phase might be replayed (only at-least-once semantic is promised). For example updating an already updated value in snapshot, or deleting an already deleted entry in snapshot. These replayed change log events should be handled specially.
mongodb.ssl.enabled	optional	false	Boolean	Whether the connector will use SSL to connect to MongoDB instances.
mongodb.ssl.invalid.hostname.allowed	optional	false	Boolean	When SSL is enabled, this setting controls whether strict hostname checking is disabled during the connection phase. If true, the connection will not prevent man-in-the-middle attacks.
mongodb.ssl.keystore	optional	(none)	String	The location of the key store file. This is optional and can be used for two-way authentication between the client and the MongoDB server.
mongodb.ssl.keystore.password	optional	(none)	String	The password for the key store file. This is optional and only needed if mongodb.ssl.keystore is configured.
mongodb.ssl.keystore.type	optional	PKCS12	String	The type of key store file. This is optional and only needed if mongodb.ssl.keystore is configured. Defaults to PKCS12.
mongodb.ssl.truststore	optional	(none)	String	The location of the trust store file for the server certificate verification.
mongodb.ssl.truststore.password	optional	(none)	String	The password for the trust store file. Used to check the integrity of the truststore, and unlock the truststore.
mongodb.ssl.truststore.type	optional	PKCS12	String	The type of trust store file. This is optional and only needed if mongodb.ssl.truststore is configured. Defaults to PKCS12.

Note: heartbeat.interval.ms is highly recommended setting a proper value larger than 0 if the collection changes slowly. The heartbeat event can push the resumeToken forward to avoid resumeToken being expired when we recover the Flink job from a checkpoint or savepoint.

Note: The mongodb.ssl.* options only take effect when the incremental snapshot is enabled (scan.incremental.snapshot.enabled = true). They are ignored by the non-incremental source.

Available Metadata

The following format metadata can be exposed as read-only (VIRTUAL) columns in a table definition.

Key	DataType	Description
database_name	STRING NOT NULL	Name of the database that contain the row.
collection_name	STRING NOT NULL	Name of the collection that contain the row.
op_ts	TIMESTAMP_LTZ(3) NOT NULL	It indicates the time that the change was made in the database. If the record is read from snapshot of the table instead of the change stream, the value is always 0.
row_kind	STRING NOT NULL	It indicates the row kind of the changelog,Note: The downstream SQL operator may fail to compare due to this new added column when processing the row retraction if the source operator chooses to output the 'row_kind' column for each record. It is recommended to use this metadata column only in simple synchronization jobs. '+I' means INSERT message, '-D' means DELETE message, '-U' means UPDATE_BEFORE message and '+U' means UPDATE_AFTER message.

The extended CREATE TABLE example demonstrates the syntax for exposing these metadata fields:

CREATE TABLE products (
    db_name         STRING METADATA FROM 'database_name' VIRTUAL,
    collection_name STRING METADATA  FROM 'collection_name' VIRTUAL,
    operation_ts    TIMESTAMP_LTZ(3) METADATA FROM 'op_ts' VIRTUAL,
    operation       STRING METADATA FROM 'row_kind' VIRTUAL,
    _id             STRING, // must be declared
    name            STRING,
    weight          DECIMAL(10,3),
    tags            ARRAY<STRING>, -- array
    price           ROW<amount DECIMAL(10,2), currency STRING>, -- embedded document
    suppliers       ARRAY<ROW<name STRING, address STRING>>, -- embedded documents
    PRIMARY KEY(_id) NOT ENFORCED
) WITH (
    'connector' = 'mongodb-cdc',
    'hosts' = 'localhost:27017,localhost:27018,localhost:27019',
    'username' = 'flinkuser',
    'password' = 'flinkpw',
    'database' = 'inventory',
    'collection' = 'products'
);

Features

Exactly-Once Processing

The MongoDB CDC connector is a Flink Source connector which will read database snapshot first and then continues to read change stream events with exactly-once processing even failures happen.

Startup Reading Position

The config option scan.startup.mode specifies the startup mode for MongoDB CDC consumer. The valid enumerations are:

• initial (default): Performs an initial snapshot on the monitored database tables upon first startup, and continue to read the latest oplog.
• latest-offset: Never to perform snapshot on the monitored database tables upon first startup, just read from the end of the oplog which means only have the changes since the connector was started.
• timestamp: Skip snapshot phase and start reading oplog events from a specific timestamp.

For example in DataStream API:

MongoDBSource.builder()
    .startupOptions(StartupOptions.latest()) // Start from latest offset
    .startupOptions(StartupOptions.timestamp(1667232000000L) // Start from timestamp
    .build()

and with SQL:

CREATE TABLE mongodb_source (...) WITH (
    'connector' = 'mongodb-cdc',
    'scan.startup.mode' = 'latest-offset', -- Start from latest offset
    ...
    'scan.startup.mode' = 'timestamp', -- Start from timestamp
    'scan.startup.timestamp-millis' = '1667232000000' -- Timestamp under timestamp startup mode
    ...
)

Snapshot Data Filters

The config option initial.snapshotting.pipeline describing the filters when copying existing data.
This can filter only required data and improve the use of indexes by the copying manager.

In the following example, the $match aggregation operator ensures that only documents in which the closed field is set to "false" are copied.

'initial.snapshotting.pipeline' = '[ { "$match": { "closed": "false" } } ]'

Change Streams

We integrate the MongoDB's official Kafka Connector to read snapshot or change events from MongoDB and drive it by Debezium's EmbeddedEngine.

Debezium's EmbeddedEngine provides a mechanism for running a single Kafka Connect SourceConnector within an application's process, and it can drive any standard Kafka Connect SourceConnector properly even which is not provided by Debezium.

We choose MongoDB's official Kafka Connector instead of the Debezium's MongoDB Connector because they use a different change data capture mechanism.

• For Debezium's MongoDB Connector, it reads the oplog.rs collection of each replica-set's master node.
• For MongoDB's Kafka Connector, it subscribes Change Stream of MongoDB.

MongoDB's oplog.rs collection doesn't keep the changed record's update before state, so it's hard to extract the full document state by a single oplog.rs record and convert it to change log stream accepted by Flink (Insert Only, Upsert, All). Additionally, MongoDB 5 (released in July 2021) has changed the oplog format, so the current Debezium connector cannot be used with it.

Change Stream is a new feature provided by MongoDB 3.6 for replica sets and sharded clusters that allows applications to access real-time data changes without the complexity and risk of tailing the oplog.
Applications can use change streams to subscribe to all data changes on a single collection, a database, or an entire deployment, and immediately react to them.

Lookup Full Document for Update Operations is a feature provided by Change Stream which can configure the change stream to return the most current majority-committed version of the updated document. Because of this feature, we can easily collect the latest full document and convert the change log to Flink's Upsert Changelog Stream.

By the way, Debezium's MongoDB change streams exploration mentioned by DBZ-435 is on roadmap.
If it's done, we can consider integrating two kinds of source connector for users to choose.

Scan Newly Added Collections

Note: This feature is available since Flink CDC 3.1.0.

The Scan Newly Added Collections feature enables you to add new collections to monitor for existing running pipeline. The newly added collections will read their snapshot data firstly and then read their change stream automatically.

Imagine this scenario: At the beginning, a Flink job monitors collections [product, user, address], but after some days we would like the job can also monitor collections [order, custom] which contain history data, and we need the job can still reuse existing state of the job. This feature can resolve this case gracefully.

The following operations show how to enable this feature to resolve above scenario. An existing Flink job which uses MongoDB CDC Source like:

    MongoDBSource<String> mongoSource = MongoDBSource.<String>builder()
        .hosts("yourHostname:27017")
        .databaseList("db") // set captured database
        .collectionList("db.product", "db.user", "db.address") // set captured collections
        .username("yourUsername")
        .password("yourPassword")
        .scanNewlyAddedTableEnabled(true) // enable scan the newly added collections feature
        .deserializer(new JsonDebeziumDeserializationSchema()) // converts SourceRecord to JSON String
        .build();
   // your business code

If we would like to add new collections [order, custom] to an existing Flink job, we just need to update the collectionList() value of the job to include [order, custom] and restore the job from previous savepoint.

Step 1: Stop the existing Flink job with savepoint.

$ ./bin/flink stop $Existing_Flink_JOB_ID

Suspending job "cca7bc1061d61cf15238e92312c2fc20" with a savepoint.
Savepoint completed. Path: file:/tmp/flink-savepoints/savepoint-cca7bc-bb1e257f0dab

Step 2: Update the collection list option for the existing Flink job.

1. update collectionList() value.
2. build the jar of updated job.

    MongoDBSource<String> mongoSource = MongoDBSource.<String>builder()
        .hosts("yourHostname:27017")
        .databaseList("db")
        .collectionList("db.product", "db.user", "db.address", "db.order", "db.custom") // set captured collections [product, user, address, order, custom]
        .username("yourUsername")
        .password("yourPassword")
        .scanNewlyAddedTableEnabled(true) // enable scan newly added tables feature
        .deserializer(new JsonDebeziumDeserializationSchema()) // converts SourceRecord to JSON String
        .build();
   // your business code

Step 3: Restore the updated Flink job from savepoint.

$ ./bin/flink run \
      --detached \
      --from-savepoint /tmp/flink-savepoints/savepoint-cca7bc-bb1e257f0dab \
      ./FlinkCDCExample.jar

Note: Please refer to Restore the job from previous savepoint for more details.

DataStream Source

The MongoDB CDC connector can also be a DataStream source. You can create a SourceFunction as the following shows:

import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.source.SourceFunction;
import org.apache.flink.cdc.debezium.JsonDebeziumDeserializationSchema;
import org.apache.flink.cdc.connectors.mongodb.MongoDBSource;

public class MongoDBSourceExample {
    public static void main(String[] args) throws Exception {
        SourceFunction<String> sourceFunction = MongoDBSource.<String>builder()
                .hosts("localhost:27017")
                .username("flink")
                .password("flinkpw")
                .databaseList("inventory") // set captured database, support regex
                .collectionList("inventory.products", "inventory.orders") //set captured collections, support regex
                .deserializer(new JsonDebeziumDeserializationSchema())
                .build();

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        env.addSource(sourceFunction)
                .print().setParallelism(1); // use parallelism 1 for sink to keep message ordering

        env.execute();
    }
}

The MongoDB CDC incremental connector (after 2.3.0) can be used as the following shows:

import org.apache.flink.api.common.eventtime.WatermarkStrategy;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.cdc.connectors.mongodb.source.MongoDBSource;
import org.apache.flink.cdc.debezium.JsonDebeziumDeserializationSchema;

public class MongoDBIncrementalSourceExample {
    public static void main(String[] args) throws Exception {
        MongoDBSource<String> mongoSource =
                MongoDBSource.<String>builder()
                        .hosts("localhost:27017")
                        .databaseList("inventory") // set captured database, support regex
                        .collectionList("inventory.products", "inventory.orders") //set captured collections, support regex
                        .username("flink")
                        .password("flinkpw")
                        .deserializer(new JsonDebeziumDeserializationSchema())
                        .build();

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
        // enable checkpoint
        env.enableCheckpointing(3000);
        // set the source parallelism to 2
        env.fromSource(mongoSource, WatermarkStrategy.noWatermarks(), "MongoDBIncrementalSource")
                .setParallelism(2)
                .print()
                .setParallelism(1);

        env.execute("Print MongoDB Snapshot + Change Stream");
    }
}

Note:

• If database regex is used, readAnyDatabase role is required.
• The incremental snapshot feature only supports after MongoDB 4.0.

Full Changelog

MongoDB 6.0 and above supports emitting change stream events containing document before and after the change was made (aka. pre- and post-images).

• The pre-image is the document before it was replaced, updated, or deleted. There is no pre-image for an inserted document.

• The post-image is the document after it was inserted, replaced, or updated. There is no post-image for a deleted document.

MongoDB CDC could make uses of pre-image and post-images to generate full-mode changelog stream including Insert, Update Before, Update After, and Delete data rows, thereby avoiding additional ChangelogNormalize downstream node.

To enable this feature, here's some prerequisites:

• MongoDB version must be 6.0 or above;
• Enable preAndPostImages feature at the database level:

db.runCommand({
  setClusterParameter: {
    changeStreamOptions: {
      preAndPostImages: {
        expireAfterSeconds: 'off' // replace with custom image expiration time
      }
    }
  }
})

• Enable changeStreamPreAndPostImages feature for collections to be monitored:

db.runCommand({
  collMod: "<< collection name >>", 
  changeStreamPreAndPostImages: {
    enabled: true 
  } 
})

• Enable MongoDB CDC's scan.full-changelog feature:

MongoDBSource.builder()
    .scanFullChangelog(true)
    ...
    .build()

or with Flink SQL:

CREATE TABLE mongodb_source (...) WITH (
    'connector' = 'mongodb-cdc',
    'scan.full-changelog' = 'true',
    ...
)

Available Source metrics

Metrics can help understand the progress of assignments, and the following are the supported Flink metrics:

Group	Name	Type	Description
namespace.schema.table	isSnapshotting	Gauge	Weather the table is snapshotting or not
namespace.schema.table	isStreamReading	Gauge	Weather the table is stream reading or not
namespace.schema.table	numTablesSnapshotted	Gauge	The number of tables that have been snapshotted
namespace.schema.table	numTablesRemaining	Gauge	The number of tables that have not been snapshotted
namespace.schema.table	numSnapshotSplitsProcessed	Gauge	The number of splits that is being processed
namespace.schema.table	numSnapshotSplitsRemaining	Gauge	The number of splits that have not been processed
namespace.schema.table	numSnapshotSplitsFinished	Gauge	The number of splits that have been processed
namespace.schema.table	snapshotStartTime	Gauge	The time when the snapshot started
namespace.schema.table	snapshotEndTime	Gauge	The time when the snapshot ended

Notice:

1. The group name is namespace.schema.table, where namespace is the actual database name, schema is the actual schema name, and table is the actual table name.
2. For MongoDB, the namespace will be set to the default value "", and the group name will be like test_database.test_table.

Data Type Mapping

BSON short for Binary JSON is a binary-encoded serialization of JSON-like format used to store documents and make remote procedure calls in MongoDB.

Flink SQL Data Type is similar to the SQL standard’s data type terminology which describes the logical type of a value in the table ecosystem. It can be used to declare input and/or output types of operations.

In order to enable Flink SQL to process data from heterogeneous data sources, the data types of heterogeneous data sources need to be uniformly converted to Flink SQL data types.

The following is the mapping of BSON type and Flink SQL type.

BSON type	Flink SQL type
	TINYINT
	SMALLINT
Int	INT
Long	BIGINT
	FLOAT
Double	DOUBLE
Decimal128	DECIMAL(p, s)
Boolean	BOOLEAN
Date Timestamp	DATE
Date Timestamp	TIME
Date	TIMESTAMP(3) TIMESTAMP_LTZ(3)
Timestamp	TIMESTAMP(0) TIMESTAMP_LTZ(0)
String ObjectId UUID Symbol MD5 JavaScript Regex	STRING
BinData	BYTES
Object	ROW
Array	ARRAY
DBPointer	ROW<$ref STRING, $id STRING>
GeoJSON	Point : ROW<type STRING, coordinates ARRAY<DOUBLE>> Line : ROW<type STRING, coordinates ARRAY<ARRAY< DOUBLE>>> ...

Reference

• MongoDB Kafka Connector
• Change Streams
• Replication
• Sharding
• Database User Roles
• WiredTiger
• Replica set protocol
• Connection String Options
• Document Pre- and Post-Images
• BSON Types
• Flink DataTypes